RAG (Retrieval-Augmented Generation) System

A comprehensive implementation of a Retrieval-Augmented Generation (RAG) system with multiple retrieval strategies and generation capabilities.

Overview

This project implements a complete RAG system with the following components:

• Document Loading: Utilities for loading and preprocessing documents
• Vector Storage: FAISS-based vector store with sentence transformers
• Retrieval: Multiple retrieval strategies (keyword-based, vector-based, hybrid)
• Generation: Template-based response generation
• QA System: Advanced question-answering system with multiple strategies

Project Structure

RAG/
├── src/                    # Core RAG components
│   ├── __init__.py        # Package initialization
│   ├── document_loader.py # Document loading utilities
│   ├── vector_store.py    # Vector storage implementation
│   └── simple_rag.py      # Basic RAG implementation
├── examples/              # Example scripts
│   ├── basic_rag_example.py # Basic RAG demonstration
│   └── qa_system.py       # Advanced QA system
├── data/                  # Data directory
│   ├── sample_docs.txt    # Sample documents for testing
│   └── knowledge_base/    # Knowledge base directory
├── notebooks/             # Jupyter notebooks
├── tests/                 # Test files
├── requirements.txt       # Python dependencies
└── README.md             # This file

Installation

1. Clone the repository (if not already done):

   git clone <repository-url>
   cd RAG

2. Install dependencies:

   pip install -r requirements.txt

3. Verify installation:

   python -c "import sentence_transformers; print('Installation successful!')"

Quick Start

Basic RAG Example

Run the basic RAG demonstration:

python examples/basic_rag_example.py

This will:

• Load sample documents about AI and machine learning
• Demonstrate simple keyword-based retrieval
• Show vector store functionality with semantic search

Advanced QA System

Run the advanced question-answering system:

python examples/qa_system.py

This provides an interactive interface with multiple retrieval strategies.

Core Components

1. Vector Store (src/vector_store.py)

Provides vector storage and similarity search using sentence transformers and FAISS:

from src.vector_store import VectorStore

# Initialize vector store
vector_store = VectorStore()

# Add documents
documents = ["Document 1", "Document 2", "Document 3"]
vector_store.add_documents(documents)

# Search for similar documents
results = vector_store.similarity_search("your query", top_k=5)

2. Simple RAG (src/simple_rag.py)

Basic RAG implementation with retrieval and generation:

from src.simple_rag import SimpleRAG

# Initialize RAG system
rag = SimpleRAG(documents)

# Query the system
result = rag.query("What is machine learning?", top_k=3)
print(result['response'])

3. QA System (examples/qa_system.py)

Advanced question-answering system with multiple strategies:

from examples.qa_system import QASystem

# Initialize QA system
qa = QASystem(documents)

# Answer questions with different strategies
answer = qa.answer_question("What is AI?", strategy="hybrid")

Features

Retrieval Strategies

1. Simple Retrieval: Keyword-based matching using TF-IDF inspired scoring
2. Vector Retrieval: Semantic search using sentence embeddings
3. Hybrid Retrieval: Combination of both approaches for better results

Generation Styles

• Answer: Direct answer generation
• Summary: Context summarization
• Explanation: Detailed explanations

Vector Store Features

• FAISS Integration: Fast similarity search
• Sentence Transformers: High-quality embeddings
• Persistence: Save and load vector stores
• Multiple Models: Support for different embedding models

Usage Examples

Adding Your Own Documents

# Load your documents
with open('your_documents.txt', 'r') as f:
    documents = f.read().split('\n\n')

# Initialize RAG system
rag = SimpleRAG(documents)

# Query your documents
result = rag.query("Your question here")

Using Different Embedding Models

# Use a different sentence transformer model
vector_store = VectorStore(model_name="all-mpnet-base-v2")

Customizing Retrieval

# Adjust number of retrieved documents
result = rag.query("Question", top_k=5)

# Use different generation style
result = rag.query("Question", style="summary")

Configuration

Environment Variables

• SENTENCE_TRANSFORMERS_HOME: Cache directory for models
• FAISS_CACHE_DIR: Cache directory for FAISS indices

Model Options

Available sentence transformer models:

• all-MiniLM-L6-v2 (default, fast, 384 dimensions)
• all-mpnet-base-v2 (high quality, 768 dimensions)
• all-MiniLM-L12-v2 (balanced, 384 dimensions)

Testing

Run the test suite:

python -m pytest tests/

Performance Considerations

• Memory Usage: Vector stores can be memory-intensive with large document collections
• Processing Speed: Sentence transformers require initial model download
• Storage: FAISS indices can be saved to disk for reuse

Troubleshooting

Common Issues

1. Import Errors: Ensure all dependencies are installed
2. Model Download: First run may take time to download models
3. Memory Issues: Consider using smaller models for large document collections

Getting Help

• Check the example scripts for usage patterns
• Review the docstrings in source files
• Ensure your Python environment has sufficient memory

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests if applicable
5. Submit a pull request

License

This project is open source and available under the MIT License.

Acknowledgments

• Sentence Transformers library for embeddings
• FAISS for efficient similarity search
• The RAG research community for foundational concepts